Content
82%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A dense, well-organized attack reference with concrete payloads, an executable PoC, severity ratings, and remediation guidance — strong on token efficiency and actionability. The main gaps are the pseudocode blind-extraction snippet and the lack of explicit technique-ordering/error-recovery guidance.
Suggestions
Make the blind-extraction example executable by showing the actual request (e.g., an ldapsearch or requests call) instead of the undefined `response` variable.
Add a short note on error-recovery paths, e.g., what to try when wildcard payloads are escaped or the bind fails, to close the workflow validation gap.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and dense: a two-line syntax primer, terse payload comments ("→ matches first user, often admin"), and short sections that each earn their place. No general-concept padding or verbosity to trim, matching the top anchor. | 5 / 5 |
Actionability | Most guidance is executable — exact payloads ("*", "admin)((|(uid=*"), a copy-paste curl PoC, and concrete defender code with escape_filter_chars. Falls short of 5 because the blind-extraction block is pseudocode: `if "found" in response` references an undefined response with no actual request code. | 4 / 5 |
Workflow Clarity | A clear implicit sequence (spot filter concatenation → attempt bypass → verify via curl PoC / ldapsearch) with verification checkpoints and a probe-response feedback loop in blind extraction. Not 5 because technique ordering is implicit and there is no error-recovery guidance (e.g., what to try when wildcards are escaped). | 4 / 5 |
Progressive Disclosure | No bundle files exist and the ~75-line body is appropriately self-contained with clear section headers and well-signaled, one-level cross-references. Not 5 because the body exceeds the 50-line simple-skill threshold and the cross-referenced paths (skills/_corpus/..., skills/ad/SKILL.md) point outside the skill bundle, making navigation unverifiable. | 4 / 5 |
Total | 17 / 20 Passed |